DATA ANALYTICS NSSC’24

Cosmic Collision-
Analysing Asteroid Risks with Data
OVERVIEW
Asteroids are small, rocky objects that orbit the Sun, primarily found in the region between Mars
and Jupiter known as the asteroid belt. They are remnants from the early solar system, formed
over 4.6 billion years ago, and are considered planetesimals, or building blocks of planets that
never coalesced into a larger body due to gravitational disturbances, particularly from Jupiter.

Asteroid impacts have been a constant force shaping Earth's history. From minor events that
leave barely a trace to catastrophic collisions that have caused mass extinctions, these cosmic
encounters have played a significant role in our planet's evolution. Given the potential for
devastating consequences, scientists worldwide are dedicated to detecting and tracking the
asteroid impact threat. To mitigate this risk, it is crucial to identify and analyse asteroids that
could pose a hazard.

Objective of the Analysis: Data-Driven Classification of Hazardous

The primary objective of this analysis is to use data analytics and machine learning techniques
to determine the likelihood of an asteroid being hazardous to Earth based on various features
provided in the dataset. This includes examining characteristics such as the asteroid's size,
orbital parameters, velocity, and proximity to Earth's orbit.

By analysing these features, the goal is to develop predictive models that classify asteroids into
hazardous and non-hazardous categories. This classification is crucial for prioritising which
asteroids require further monitoring, potential deflection efforts, or other mitigation strategies.

We will also identify asteroids exhibiting unusual or anomalous behaviour that may warrant
closer attention.
 GENERAL INSTRUCTIONS
1. The dataset for the problem can be found here: Dataset
2. The description of the features used is given here: Description of the Features
3. The problem statement consists of 5 parts, and each part has a few subproblems.
4. These problems are based on the Solar System Dynamics data captured by NASA.
5. Students can participate in teams of 3-4 members.
6. The weightage of each question is mentioned next to the question.
7. Participants are free to use any programming language, environment, and
library. However, it is preferable to use Google Colab or Jupyter Notebook as the
programming environments.
8. For submission, participants should submit the .ipynb file of the solution. The
code should be well-commented to clearly convey the work done. Each team
should also submit a PDF report containing a detailed solution and approach. All
plots and outputs must be shown in the report with proper explanations and
descriptions. Final answers should be highlighted.
9. All code snippets should be attached to the report.
10. Only 15 minutes will be allocated for each team to present their solutions.
Therefore, all relevant conclusions must be presented promptly.
11. Participants will be evaluated on the clarity and applicability of the solution, their
innovation, and the feasibility of their conclusions.
12. Important: Ensure that each .ipynb code block is clearly associated
with the corresponding question and sub-question. Use markdown to
label each block, so it is easy to identify which problem and
subproblem the block is solving. Failure to do so may lead to
disqualification.

Note: The problem statement below is worth a total of 100 MARKS. Your
score will be scaled to 60 MARKS, with the remaining 40 MARKS allocated
for the presentation.
 PROBLEM STATEMENT
1. Exploratory Data Analysis (EDA) (20 points)

1.1 Data Inspection (7 points):

● Inspect the dataset and determine the data types of all features (numerical, categorical).
(1 MARK)
● Calculate and analyse basic statistics for each numerical feature, including range, mean,
median, standard deviation, and quartiles. (2 MARKS)
● Identify features that have missing values. (1 MARK)
● Identify the numerical and categorical features of the dataset to use for further analysis.
(1 MARK)
● Use imputation to fill the null values in the dataset. How is this process different for
numerical and categorical columns? (2 MARKS)

1.2 Statistical Inference (6 points):

● Plot the distribution of numerical features to assess the skewness of the data. Does this
dataset require normalisation? If yes, normalise/scale the dataset. (Hint: Use
histograms) (2 MARKS)
● Identify potential outliers in the numerical columns using any statistical technique (e.g.,
box plots, z-score, etc.). (2 MARKS)
● Explore the relationship between different features using scatter plots or correlation
matrices. (Hint: Use Seaborn or similar libraries) (2 MARKS)

1.3 Visualisation (4 points):

● Create a pairplot using Seaborn to visualise relationships between multiple numerical

1.4 Tackling Class Imbalance (3 points):

● Is there a classification bias (class imbalance) in this dataset? If yes, how would you
tackle it? (2 MARKS)
● Discuss the implication of class imbalance on model performance. (1 MARK)
 2. Numerical Interpretation and Mathematical Analysis (20
points)

2.1 Feature Engineering (15 points):

● Combine the approach_date, month, and year features into a single feature representing
the day of the year. Convert it into a ‘datetime’ format. (1 MARK)
● Calculate the ratio of Miss Distance vs. Semi-major axis. Create a 'Time Until Approach'
feature based on the difference between the 'Epoch Date Close Approach' and the current
date. (3 MARKS)
● Calculate the eccentricity of the orbit, average orbital velocity, and orbital period using
Kepler’s Law. (3 MARKS)
● Calculate the heliocentric distance, escape velocity, and specific orbital energy. (3
MARKS)
● Calculate the Specific Angular Momentum using the formula: h=sqrt(GMa(1−e²)). (1
MARK)
● Calculate the velocity at Perihelion and Aphelion. (1 MARK)
● Average the Miss distance of various categories and find the closest approach distance. (1
MARK)
● Calculate Synodic Period and Mean Motion using the orbital period. (2 MARKS)

2.2 Additional Features (5 points):

● Create additional features as per your understanding of the problem for improving
accuracy. More marks are awarded for innovative and effective features. (5 MARKS)

3. Handling Binned Values (5 points):

Note: Binned features are categorical variables where values are grouped into discrete
categories such as: [very slow, slow, fast, very fast, etc.].

● Modify the binned features that have an ordinal relationship in this manner:

(very slow = 0, slow = 1, fast = 2, very fast = 3, etc). (2 MARKS)

● One-hot encode the binned features whose relationship is not strictly ordinal. (3
MARKS)
4. Hazardous Classification (35 points):
● Build a robust and efficient classifier to classify asteroids as Hazardous (1) or Not
Hazardous (0). (7 MARKS)
● Implement K-Fold Cross Validation for training. Train the dataset for all values of K
from 2 to 10. Plot the loss and accuracy versus epochs for these K values. (12 MARKS)
● Optimise all the hyperparameters used in the classifier by selecting an appropriate
optimisation method. (8 MARKS)
● Plot the ROC curve and Confusion Matrix to quantify the performance of your classifier.
(4 MARKS)
● Use SHAP Values, Permutation Importance, or Partial Dependence Plots to list the most
and least useful features. (4 MARKS)

5. Anomaly Detection (20 points):